Guides

Guides each take one task from start to finish. Pick one by what you want to do.

Where to start

• You have just installed Kraft and nothing is connected. Follow Your first work item from a terminal or your agent, or The setup wizard on first start in the web UI. Then read File a work item.
• A work item is waiting at a gate for you to approve or reject, or you want to comment on one. Read Reviewing a change.
• You work in a coding agent and want to drive Kraft from it. Read Use Kraft from your agent.
• An item stopped on a cap. Read Raise a cap on a stopped item.
• You want the board on your phone. Read Kraft on a phone, then Remote access.
• You are upgrading Kraft, backing it up, or going back a version. Read Upgrade Kraft, Back up and restore the database or Roll back to 1.4.
• You want to change the chain, a harness or the server. Pick from the groups below.

In this section

Day to day

• The setup wizard on first start: connect your first repo with the setup steps the board shows when none is connected.
• File a work item: file a work item from the board's composer or the full new-item page.
• Use Kraft from your agent: drive Kraft with /kraft:* slash commands from a coding agent.
• Reviewing a change: approve or reject the gate a work item waits at, in the web UI or the CLI, and comment on it at any point.
• Raise a cap on a stopped item: raise the cap that stopped a work item from the interface or the command line, and retry it.
• Kraft for VS Code: clear gates, review a branch and edit config files from the editor.
• Kraft on a phone: the phone layout, its bottom bar, the board's cards and the buttons on a stopped item.
• Get notified when an item needs you: send a webhook or show a browser alert when an item waits at a gate or stops, and test it.

Customize and automate

• Write your own chain: add a lint node with an agent that fixes what it reports, a time cap, and no skipping.
• Add a security review or a gate reviewer: put the shipped security-review and gate-review skills into a chain of your own.
• Schedule or webhook work: file paused work items from a cron schedule or an HTTP call.

Agents and harnesses

• Switch a task to another harness: run agent tasks on Codex or another CLI instead of Claude Code.
• Add or override a harness: add an agent CLI with a YAML file.
• Build a worker Kit: build a Docker Sandbox Kit for Claude workers and run a repository's sandbox from it.
• Give a worker its login: give the agent CLI a worker runs on a login it can use headless, and check that it works.

Run the server

• Remote access: reach the board from a phone or another machine, over Tailscale, your local network or a tunnel, with a password.
• Put a proxy in front of Kraft: let a tunnel or reverse proxy on this machine pass webhooks and scripts to a loopback-bound Kraft.
• Restart Kraft without ending running agents: pause the items that have an agent running, restart the server, and resume them.
• Back up and restore the database: copy the database while the server runs, restore it, and use the copy Kraft makes before an upgrade.
• Logs and disk space: find the logs, and reclaim the space worktrees use.
• Run Kraft as a service: start Kraft at login, and run a second instance.
• Upgrade Kraft: back up, upgrade and check an installed Kraft, pin a version, and roll back to an earlier 2.x release.
• Upgrade your configuration: take new shipped chains, library tasks and policy after an upgrade without losing your edits.
• Upgrade from 1.4: what 2.0 changes from 1.4, and the steps to update from it.
• Upgrade from 0.x or a release candidate: move from a 1.5.0 release candidate, an install from before the kraft-sdlc rename, or a 0.x home with the old template configuration.
• Roll back to 1.4: what a rollback does to the database and config, and the steps from 2.0 back to 1.4.

Without a server. Kraft Lite: run a chain inside one agent session with no service.